This standard outlines a common set of status codes in a similar vein to HTTP statuses. This provides a shared set of signals to allow smart contracts to react to situations autonomously, expose localized error messages to users, and so on.
The current state of the art is to either `revert` on anything other than a clear success (ie: require human intervention), or return a low-context `true` or `false`. Status codes are similar-but-orthogonal to `revert`ing with a reason, but aimed at automation, debugging, and end-user feedback (including translation). _They are fully compatible with both `revert` and `revert`-with-reason._
As is the case with HTTP, having a standard set of known codes has many benefits for developers. They remove friction from needing to develop your own schemes for every contract, makes inter-contract automation easier, and makes it easier to broadly understand which of the finite states your request produced. Importantly, it makes it much easier to distinguish between expected errors states, truly exceptional conditions that require halting execution, normal state transitions, and various success cases.
HTTP status codes are widely used for this purpose. BEAM languages use atoms and tagged tuples to signify much the same information. Both provide a lot of information both to the programmer (debugging for instance), and to the program that needs to decide what to do next.
Status codes convey a much richer set of information [than Booleans](https://existentialtype.wordpress.com/2011/03/15/boolean-blindness/), and are able to be reacted to autonomously unlike arbitrary strings.
### User Experience (UX)
_End users get little to no feedback, and there is no translation layer._
Since ERC1066 status codes are finite and known in advance, we can leverage [ERC1444](https://eips.ethereum.org/EIPS/eip-1444) to provide global, human-readable sets of status messages. These may also be translated into any language, differing levels of technical detail, added as `revert` messages, natspecs, and so on.
At time of writing, other than stepping through EVM execution and inspecting memory dumps directly, it is very difficult to understand what is happening during smart contract execution. By returning more context, developers can write well-decomposed tests and assert certain codes are returned as an expression of where the smart contract got to. This includes status codes as bare values, `event`s, and `revert`s.
Having a fixed set of codes also makes it possible to write common helper functions to react in common ways to certain signals. This can live off- or on-chain library, lowering the overhead in building smart contracts, and helping raise code quality with trusted shared components.
We also see a desire for this [in transactions](http://eips.ethereum.org/EIPS/eip-658), and there's no reason that these status codes couldn't be used by the EVM itself.
_Smart contracts don’t know much about the result of a request beyond pass/fail; they can be smarter with more context._
Smart contracts are largely intended to be autonomous. While each contract may define a specific interface, having a common set of semantic codes can help developers write code that can react appropriately to various situations.
While clearly related, status codes are complementary to `revert`-with-reason. Status codes are not limited to rolling back the transaction, and may represent known error states without halting execution. They may also represent off-chain conditions, supply a string to revert, signal time delays, and more.
All of this enables contracts to share a common vocabulary of state transitions, results, and internal changes, without having to deeply understand custom status enums or the internal business logic of collaborator contracts.
Codes break nicely into a 16x16 matrix, represented as a 2-digit hex number. The high nibble represents the code's kind or "category", and the low nibble contains the state or "reason". We present them below as separate tables per range for explanatory and layout reasons.
Contracts may have special states that they need to signal. This proposal only outlines the broadest meanings, but implementers may have very specific meanings for each, as long as they are coherent with the broader definition.
| `0x*F` | `0x0F` Informational or Metadata | `0x1F` Permission Details or Control Conditions | `0x2F` Matching Meta or Info | `0x3F` Negotiation Rules or Participation Info | `0x4F` Availability Rules or Info (ex. time since or until) | `0x5F` Token or Financial Information | `0x6F` [reserved] | `0x7F` [reserved] | `0x8F` [reserved] | `0x9F` [reserved] | `0xAF` App-Specific Meta or Info | `0xBF` [reserved] | `0xCF` [reserved] | `0xDF` [reserved] | `0xEF` Cryptography, ID, or Proof Metadata | `0xFF` Off-Chain Info or Meta |
Status codes are encoded as a `byte`. Hex values break nicely into high and low nibbles: `category` and `reason`. For instance, `0x01` stands for general success (ie: `true`) and `0x00` for general failure (ie: `false`).
As a general approach, all even numbers are blocking conditions (where the receiver does not have control), and odd numbers are nonblocking (the receiver is free to contrinue as they wish). This aligns both a simple bit check with the common encoding of Booleans.
`uint8` feels even more similar to HTTP status codes, and enums don't require as much casting. However does not break as evenly as a square table (256 doesn't look as nice in base 10).
Packing multiple codes into a single `bytes32` is nice in theory, but poses additional challenges. Unused space may be interpeted as `0x00 Failure`, you can only efficiently pack four codes at once, and there is a challenge in ensuring that code combinations are sensible. Forcing four codes into a packed representation encourages multiple status codes to be returned, which is often more information than strictly necessarily. This can lead to paradoxical results (ex `0x00` and `0x01` together), or greater resorces allocated to interpreting 256<sup>4</sup> (4.3 billion) permutations.
While there may be cases where packing a byte array of status codes may make sense, the simplest, most forwards-compatible method of transmission is as the first value of a multiple return.
Familiarity is also a motivating factor. A consistent position and encoding together follow the principle of least surprise. It is both viewable as a "header" in the HTTP analogy, or like the "tag" in BEAM tagged tuples.
Developers should not be required to memorize 256 codes. However, they break nicely into a table. Cognitive load is lowered by organizing the table into categories and reasons. `0x10` and `0x11` belong to the same category, and `0x04` shares a reason with `0x24`
While this repository includes helper enums, we have found working directly in the hex values to be quite natural. Status code `0x10` is just as comfortable as HTTP 401, for example.
One commonly requested application of this spec is human-readable translations of codes. This has been moved to its own proposal: [ERC-1444](https://github.com/ethereum/EIPs/pull/1444/), primarily due to a desire to keep both specs focused.
The `0xA` category is reserved for application-specific statuses. In the case that 256 codes become insufficient, `bytes1` may be embedded in larger byte arrays.
The EVM also returns a status code in transactions; specifically `0x00` and `0x01`. This proposal both matches the meanings of those two codes, and could later be used at the EVM level.
Much like how HTTP status codes have large unused ranges, there are totally empty sections in this proposal. The intent is to not impose a complete set of codes up front, and to allow users to suggest uses for these spaces as time progresses.
### Beyond Errors
This spec is intended to be much more than a set of common errors. One design goal is to enable easier contract-to-contract communication, protocols built on top of status codes, and flows that cross off-chain. Many of these cases include either expected kinds of exception state (as opposed to true errors), neutral states, time logic, and various successes.
Just like how HTTP 200 has a different meaning from HTTP 201, ERC-1066 status codes can relay information between contract beyond simply pass or fail. They can be thought of as the edges in a graph that has smart contracts as nodes.
### Fully `revert`able
_This spec is fully compatible with `revert`-with-reason and does not intend to supplant it in any way._ Both by reverting with a common code, the developer can determine what went wrong from a set of known error states.
Further, by leveraging ERC-1066 and a translation table (such as in ERC-1444) in conjunction, developers and end users alike can receive fully automated human-readable error messages in the language and phrasing of their choice.
Nibble order makes no difference to the machine, and is purely mnemonic. This design was originally in opposite order, but changed it for a few convenience factors. Since it's a different scheme from HTTP, it may feel strange initially, but becomes very natural after a couple hours of use.
